Purpose

Vascular Ehlers-Danlos syndrome (vEDS), caused by COL3A1 pathogenic variants, is a rare heritable aortic and arterial disorder associated with early mortality, mainly due to spontaneous vascular dissections and ruptures. Improved methods for diagnosing vEDS are needed so that guideline-based management can be initiated to prevent deadly complications and differentiate vEDS from overlapping conditions like hypermobile EDS (hEDS).

Methods

We implemented an AI facial recognition model based on the PhenoScore framework using a support vector machine (SVM) trained on facial images of thirty individuals, aged 6-65 years, with vEDS from the Montalcino Aortic Consortium (MAC), control images from the Chicago Face Database (CFD), and publicly available images of individuals with hEDS. Cross- validation was used to train the SVM, and statistical measures to evaluate the model performance were calculated. Local Interpretable Model-agnostic Explanations (LIME) was used to generate facial heatmaps highlighting the features driving the model’s predictions.

Results

The AI classifier showed excellent performance with as few as thirteen vEDS training images and distinguished vEDS from both controls and individuals with hEDS with high accuracy, achieving an area under the receiver operating characteristic curve (AUC) ≥ 0.97. LIME highlighted facial regions already established to characterize the facial features of vEDS patients (e.g., prominent eyes).

Conclusion

Our results demonstrate the potential of AI-based facial analysis for diagnosing vEDS. This method democratizes the early diagnosis of vEDS by reducing dependence on genetic testing, enabling optimal management and improved outcomes, particularly in resource-limited areas.